Commit | Line | Data |
---|---|---|
c906108c SS |
1 | # Generate the main loop of the simulator. |
2 | # Copyright (C) 1996, 1997, 1998, 1999 Free Software Foundation, Inc. | |
3 | # Contributed by Cygnus Support. | |
4 | # | |
5 | # This file is part of the GNU simulators. | |
6 | # | |
7 | # This program is free software; you can redistribute it and/or modify | |
8 | # it under the terms of the GNU General Public License as published by | |
9 | # the Free Software Foundation; either version 2, or (at your option) | |
10 | # any later version. | |
11 | # | |
12 | # This program is distributed in the hope that it will be useful, | |
13 | # but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | # GNU General Public License for more details. | |
16 | # | |
17 | # You should have received a copy of the GNU General Public License along | |
18 | # with this program; if not, write to the Free Software Foundation, Inc., | |
19 | # 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
20 | # | |
21 | # This file creates two files: eng.hin and mloop.cin. | |
22 | # eng.hin defines a few macros that specify what kind of engine was selected | |
23 | # based on the arguments to this script. | |
24 | # mloop.cin contains the engine. | |
25 | # | |
26 | # ??? Rename mloop.c to eng.c? | |
27 | # ??? Rename mainloop.in to engine.in? | |
28 | # ??? Add options to specify output file names? | |
29 | # ??? Rename this file to genengine.sh? | |
30 | # | |
31 | # Syntax: genmloop.sh [options] | |
32 | # | |
33 | # Options: | |
34 | # | |
35 | # -mono | -multi | |
36 | # - specify single cpu or multiple cpus (number specifyable at runtime), | |
37 | # maximum number is a configuration parameter | |
38 | # - -multi wip | |
39 | # | |
40 | # -fast: include support for fast execution in addition to full featured mode | |
41 | # | |
42 | # Full featured mode is for tracing, profiling, etc. and is always | |
43 | # provided. Fast mode contains no frills, except speed. | |
44 | # A target need only provide a "full" version of one of | |
45 | # simple,scache,pbb. If the target wants it can also provide a fast | |
46 | # version of same. It can't provide more than this. | |
47 | # ??? Later add ability to have another set of full/fast semantics | |
48 | # for use in with-devices/with-smp situations (pbb can be inappropriate | |
49 | # here). | |
50 | # | |
51 | # -full-switch: same as -fast but for full featured version of -switch | |
52 | # Only needed if -fast present. | |
53 | # | |
54 | # -simple: simple execution engine (the default) | |
55 | # | |
56 | # This engine fetches and executes one instruction at a time. | |
57 | # Field extraction is done in the semantic routines. | |
58 | # | |
59 | # ??? There are two possible flavours of -simple. One that extracts | |
60 | # fields in the semantic routine (which is what is implemented here), | |
61 | # and one that stores the extracted fields in ARGBUF before calling the | |
62 | # semantic routine. The latter is essentially the -scache case with a | |
63 | # cache size of one (and the scache lookup code removed). There are no | |
64 | # current uses of this and it's not clear when doing this would be a win. | |
65 | # More complicated ISA's that want to use -simple may find this a win. | |
66 | # Should this ever be desirable, implement a new engine style here and | |
67 | # call it -extract (or some such). It's believed that the CGEN-generated | |
68 | # code for the -scache case would be usable here, so no new code | |
69 | # generation option would be needed for CGEN. | |
70 | # | |
71 | # -scache: use the scache to speed things up (not always a win) | |
72 | # | |
73 | # This engine caches the extracted instruction before executing it. | |
74 | # When executing instructions they are first looked up in the scache. | |
75 | # | |
76 | # -pbb: same as -scache but extract a (pseudo-) basic block at a time | |
77 | # | |
78 | # This engine is basically identical to the scache version except that | |
79 | # extraction is done a pseudo-basic-block at a time and the address of | |
80 | # the scache entry of a branch target is recorded as well. | |
81 | # Additional speedups are then possible by defering Ctrl-C checking | |
82 | # to the end of basic blocks and by threading the insns together. | |
83 | # We call them pseudo-basic-block's instead of just basic-blocks because | |
84 | # they're not necessarily basic-blocks, though normally are. | |
85 | # | |
86 | # -parallel-read: support parallel execution with read-before-exec support. | |
87 | # -parallel-write: support parallel execution with write-after-exec support. | |
88 | # | |
89 | # One of these options is specified in addition to -simple, -scache, | |
90 | # -pbb. Note that while the code can determine if the cpu supports | |
91 | # parallel execution with HAVE_PARALLEL_INSNS [and thus this option is | |
92 | # technically unnecessary], having this option cuts down on the clutter | |
93 | # in the result. | |
94 | # | |
95 | # -switch file: specify file containing semantics implemented as a switch() | |
96 | # | |
97 | # -cpu <cpu-family> | |
98 | # | |
99 | # Specify the cpu family name. | |
100 | # | |
101 | # -infile <input-file> | |
102 | # | |
103 | # Specify the mainloop.in input file. | |
104 | # | |
105 | # Only one of -scache/-pbb may be selected. | |
106 | # -simple is the default. | |
107 | # | |
108 | #### | |
109 | # | |
110 | # TODO | |
111 | # - build mainloop.in from .cpu file | |
112 | ||
113 | type=mono | |
114 | #scache= | |
115 | #fast= | |
116 | #full_switch= | |
117 | #pbb= | |
118 | parallel=no | |
119 | switch= | |
120 | cpu="unknown" | |
121 | infile="" | |
122 | ||
123 | while test $# -gt 0 | |
124 | do | |
125 | case $1 in | |
126 | -mono) type=mono ;; | |
127 | -multi) type=multi ;; | |
128 | -no-fast) ;; | |
129 | -fast) fast=yes ;; | |
130 | -full-switch) full_switch=yes ;; | |
131 | -simple) ;; | |
132 | -scache) scache=yes ;; | |
133 | -pbb) pbb=yes ;; | |
134 | -no-parallel) ;; | |
135 | -parallel-read) parallel=read ;; | |
136 | -parallel-write) parallel=write ;; | |
137 | -switch) shift ; switch=$1 ;; | |
138 | -cpu) shift ; cpu=$1 ;; | |
139 | -infile) shift ; infile=$1 ;; | |
140 | *) echo "unknown option: $1" >&2 ; exit 1 ;; | |
141 | esac | |
142 | shift | |
143 | done | |
144 | ||
145 | # Argument validation. | |
146 | ||
147 | if [ x$scache = xyes -a x$pbb = xyes ] ; then | |
148 | echo "only one of -scache and -pbb may be selected" >&2 | |
149 | exit 1 | |
150 | fi | |
151 | ||
152 | if [ "x$cpu" = xunknown ] ; then | |
153 | echo "cpu family not specified" >&2 | |
154 | exit 1 | |
155 | fi | |
156 | ||
157 | if [ "x$infile" = x ] ; then | |
158 | echo "mainloop.in not specified" >&2 | |
159 | exit 1 | |
160 | fi | |
161 | ||
162 | lowercase='abcdefghijklmnopqrstuvwxyz' | |
163 | uppercase='ABCDEFGHIJKLMNOPQRSTUVWXYZ' | |
164 | CPU=`echo ${cpu} | tr "${lowercase}" "${uppercase}"` | |
165 | ||
166 | ########################################################################## | |
167 | ||
168 | rm -f eng.hin | |
169 | exec 1>eng.hin | |
170 | ||
171 | echo "/* engine configuration for ${cpu} */" | |
172 | echo "" | |
173 | ||
174 | echo "/* WITH_FAST: non-zero if a fast version of the engine is available" | |
175 | echo " in addition to the full-featured version. */" | |
176 | if [ x$fast = xyes ] ; then | |
177 | echo "#define WITH_FAST 1" | |
178 | else | |
179 | echo "#define WITH_FAST 0" | |
180 | fi | |
181 | ||
182 | echo "" | |
183 | echo "/* WITH_SCACHE_PBB_${CPU}: non-zero if the pbb engine was selected. */" | |
184 | if [ x$pbb = xyes ] ; then | |
185 | echo "#define WITH_SCACHE_PBB_${CPU} 1" | |
186 | else | |
187 | echo "#define WITH_SCACHE_PBB_${CPU} 0" | |
188 | fi | |
189 | ||
190 | echo "" | |
191 | echo "/* HAVE_PARALLEL_INSNS: non-zero if cpu can parallelly execute > 1 insn. */" | |
192 | if [ x$parallel != xno ] ; then | |
193 | echo "#define HAVE_PARALLEL_INSNS 1" | |
194 | if [ x$parallel = xread ] ; then | |
195 | echo "/* Parallel execution is supported by read-before-exec. */" | |
196 | echo "#define WITH_PARALLEL_READ 1" | |
197 | echo "#define WITH_PARALLEL_WRITE 0" | |
198 | else | |
199 | echo "/* Parallel execution is supported by write-after-exec. */" | |
200 | echo "#define WITH_PARALLEL_READ 0" | |
201 | echo "#define WITH_PARALLEL_WRITE 1" | |
202 | fi | |
203 | else | |
204 | echo "#define HAVE_PARALLEL_INSNS 0" | |
205 | echo "#define WITH_PARALLEL_READ 0" | |
206 | echo "#define WITH_PARALLEL_WRITE 0" | |
207 | fi | |
208 | ||
209 | if [ "x$switch" != x ] ; then | |
210 | echo "" | |
211 | echo "/* WITH_SEM_SWITCH_FULL: non-zero if full-featured engine is" | |
212 | echo " implemented as a switch(). */" | |
213 | if [ x$fast != xyes -o x$full_switch = xyes ] ; then | |
214 | echo "#define WITH_SEM_SWITCH_FULL 1" | |
215 | else | |
216 | echo "#define WITH_SEM_SWITCH_FULL 0" | |
217 | fi | |
218 | echo "" | |
219 | echo "/* WITH_SEM_SWITCH_FAST: non-zero if fast engine is" | |
220 | echo " implemented as a switch(). */" | |
221 | if [ x$fast = xyes ] ; then | |
222 | echo "#define WITH_SEM_SWITCH_FAST 1" | |
223 | else | |
224 | echo "#define WITH_SEM_SWITCH_FAST 0" | |
225 | fi | |
226 | fi | |
227 | ||
228 | # Decls of functions we define. | |
229 | ||
230 | echo "" | |
231 | echo "/* Functions defined in the generated mainloop.c file" | |
232 | echo " (which doesn't necessarily have that file name). */" | |
233 | echo "" | |
234 | echo "extern ENGINE_FN ${cpu}_engine_run_full;" | |
235 | echo "extern ENGINE_FN ${cpu}_engine_run_fast;" | |
236 | ||
237 | if [ x$pbb = xyes ] ; then | |
238 | echo "" | |
239 | echo "extern SEM_PC ${cpu}_pbb_begin (SIM_CPU *, int);" | |
240 | echo "extern SEM_PC ${cpu}_pbb_chain (SIM_CPU *, SEM_ARG);" | |
241 | echo "extern SEM_PC ${cpu}_pbb_cti_chain (SIM_CPU *, SEM_ARG, SEM_PC *, PCADDR);" | |
242 | echo "extern void ${cpu}_pbb_before (SIM_CPU *, SCACHE *);" | |
243 | echo "extern void ${cpu}_pbb_after (SIM_CPU *, SCACHE *);" | |
244 | fi | |
245 | ||
246 | ########################################################################## | |
247 | ||
248 | rm -f tmp-mloop.cin mloop.cin | |
249 | exec 1>tmp-mloop.cin | |
250 | ||
251 | # We use @cpu@ instead of ${cpu} because we still need to run sed to handle | |
252 | # transformation of @cpu@ for mainloop.in, so there's no need to use ${cpu} | |
253 | # here. | |
254 | ||
255 | cat << EOF | |
256 | /* This file is generated by the genmloop script. DO NOT EDIT! */ | |
257 | ||
258 | /* Enable switch() support in cgen headers. */ | |
259 | #define SEM_IN_SWITCH | |
260 | ||
261 | #define WANT_CPU @cpu@ | |
262 | #define WANT_CPU_@CPU@ | |
263 | ||
264 | #include "sim-main.h" | |
265 | #include "bfd.h" | |
266 | #include "cgen-mem.h" | |
267 | #include "cgen-ops.h" | |
268 | #include "sim-assert.h" | |
269 | ||
270 | /* Fill in the administrative ARGBUF fields required by all insns, | |
271 | virtual and real. */ | |
272 | ||
273 | static INLINE void | |
274 | @cpu@_fill_argbuf (const SIM_CPU *cpu, ARGBUF *abuf, const IDESC *idesc, | |
275 | PCADDR pc, int fast_p) | |
276 | { | |
277 | #if WITH_SCACHE | |
278 | SEM_SET_CODE (abuf, idesc, fast_p); | |
279 | ARGBUF_ADDR (abuf) = pc; | |
280 | #endif | |
281 | ARGBUF_IDESC (abuf) = idesc; | |
282 | } | |
283 | ||
284 | /* Fill in tracing/profiling fields of an ARGBUF. */ | |
285 | ||
286 | static INLINE void | |
287 | @cpu@_fill_argbuf_tp (const SIM_CPU *cpu, ARGBUF *abuf, | |
288 | int trace_p, int profile_p) | |
289 | { | |
290 | ARGBUF_TRACE_P (abuf) = trace_p; | |
291 | ARGBUF_PROFILE_P (abuf) = profile_p; | |
292 | } | |
293 | ||
294 | #if WITH_SCACHE_PBB | |
295 | ||
296 | /* Emit the "x-before" handler. | |
297 | x-before is emitted before each insn (serial or parallel). | |
298 | This is as opposed to x-after which is only emitted at the end of a group | |
299 | of parallel insns. */ | |
300 | ||
301 | static INLINE void | |
302 | @cpu@_emit_before (SIM_CPU *current_cpu, SCACHE *sc, PCADDR pc, int first_p) | |
303 | { | |
304 | ARGBUF *abuf = &sc[0].argbuf; | |
305 | const IDESC *id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEFORE]; | |
306 | ||
307 | abuf->fields.before.first_p = first_p; | |
308 | @cpu@_fill_argbuf (current_cpu, abuf, id, pc, 0); | |
309 | /* no need to set trace_p,profile_p */ | |
310 | } | |
311 | ||
312 | /* Emit the "x-after" handler. | |
313 | x-after is emitted after a serial insn or at the end of a group of | |
314 | parallel insns. */ | |
315 | ||
316 | static INLINE void | |
317 | @cpu@_emit_after (SIM_CPU *current_cpu, SCACHE *sc, PCADDR pc) | |
318 | { | |
319 | ARGBUF *abuf = &sc[0].argbuf; | |
320 | const IDESC *id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_AFTER]; | |
321 | ||
322 | @cpu@_fill_argbuf (current_cpu, abuf, id, pc, 0); | |
323 | /* no need to set trace_p,profile_p */ | |
324 | } | |
325 | ||
326 | #endif /* WITH_SCACHE_PBB */ | |
327 | ||
328 | EOF | |
329 | ||
330 | ${SHELL} $infile support | |
331 | ||
332 | ########################################################################## | |
333 | ||
334 | # Simple engine: fetch an instruction, execute the instruction. | |
335 | # | |
336 | # Instruction fields are not extracted into ARGBUF, they are extracted in | |
337 | # the semantic routines themselves. However, there is still a need to pass | |
338 | # and return misc. information to the semantic routines so we still use ARGBUF. | |
339 | # [One could certainly implement things differently and remove ARGBUF. | |
340 | # It's not clear this is necessarily always a win.] | |
341 | # ??? The use of the SCACHE struct is for consistency with the with-scache | |
342 | # case though it might be a source of confusion. | |
343 | ||
344 | if [ x$scache != xyes -a x$pbb != xyes ] ; then | |
345 | ||
346 | cat << EOF | |
347 | ||
348 | #define FAST_P 0 | |
349 | ||
350 | void | |
351 | @cpu@_engine_run_full (SIM_CPU *current_cpu) | |
352 | { | |
353 | #define FAST_P 0 | |
354 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
355 | /* ??? Use of SCACHE is a bit of a hack as we don't actually use the scache. | |
356 | We do however use ARGBUF so for consistency with the other engine flavours | |
357 | the SCACHE type is used. */ | |
358 | SCACHE cache[MAX_LIW_INSNS]; | |
359 | SCACHE *sc = &cache[0]; | |
360 | ||
361 | EOF | |
362 | ||
363 | if [ x$parallel != xno ] ; then | |
364 | cat << EOF | |
365 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
366 | PAREXEC *par_exec; | |
367 | ||
368 | EOF | |
369 | fi | |
370 | ||
371 | # Any initialization code before looping starts. | |
372 | # Note that this code may declare some locals. | |
373 | ${SHELL} $infile init | |
374 | ||
375 | if [ x$parallel != xno ] ; then | |
376 | cat << EOF | |
377 | ||
378 | #if defined (HAVE_PARALLEL_EXEC) && defined (__GNUC__) | |
379 | { | |
380 | if (! CPU_IDESC_READ_INIT_P (current_cpu)) | |
381 | { | |
382 | /* ??? Later maybe paste read.c in when building mainloop.c. */ | |
383 | #define DEFINE_LABELS | |
384 | #include "readx.c" | |
385 | CPU_IDESC_READ_INIT_P (current_cpu) = 1; | |
386 | } | |
387 | } | |
388 | #endif | |
389 | ||
390 | EOF | |
391 | fi | |
392 | ||
393 | cat << EOF | |
394 | ||
395 | #if WITH_SEM_SWITCH_FULL && defined (__GNUC__) | |
396 | { | |
397 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
398 | { | |
399 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
400 | #define DEFINE_LABELS | |
401 | #include "$switch" | |
402 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
403 | } | |
404 | } | |
405 | #endif | |
406 | ||
407 | do | |
408 | { | |
409 | /* begin full-exec-simple */ | |
410 | EOF | |
411 | ||
412 | ${SHELL} $infile full-exec-simple | |
413 | ||
414 | cat << EOF | |
415 | /* end full-exec-simple */ | |
416 | ||
417 | ++ CPU_INSN_COUNT (current_cpu); | |
418 | } | |
419 | while (0 /*CPU_RUNNING_P (current_cpu)*/); | |
420 | } | |
421 | ||
422 | #undef FAST_P | |
423 | ||
424 | EOF | |
425 | ||
426 | #################################### | |
427 | ||
428 | # Simple engine: fast version. | |
429 | # ??? A somewhat dubious effort, but for completeness' sake. | |
430 | ||
431 | if [ x$fast = xyes ] ; then | |
432 | ||
433 | cat << EOF | |
434 | ||
435 | #define FAST_P 1 | |
436 | ||
437 | FIXME: "fast simple version unimplemented, delete -fast arg to genmloop.sh." | |
438 | ||
439 | #undef FAST_P | |
440 | ||
441 | EOF | |
442 | ||
443 | fi # -fast | |
444 | ||
445 | fi # simple engine | |
446 | ||
447 | ########################################################################## | |
448 | ||
449 | # Scache engine: lookup insn in scache, fetch if missing, then execute it. | |
450 | ||
451 | if [ x$scache = xyes ] ; then | |
452 | ||
453 | cat << EOF | |
454 | ||
455 | static INLINE SCACHE * | |
456 | @cpu@_scache_lookup (SIM_CPU *current_cpu, PCADDR vpc, SCACHE *scache, | |
457 | unsigned int hash_mask, int FAST_P) | |
458 | { | |
459 | /* First step: look up current insn in hash table. */ | |
460 | SCACHE *sc = scache + SCACHE_HASH_PC (vpc, hash_mask); | |
461 | ||
462 | /* If the entry isn't the one we want (cache miss), | |
463 | fetch and decode the instruction. */ | |
464 | if (sc->argbuf.addr != vpc) | |
465 | { | |
466 | insn_t insn; | |
467 | ||
468 | if (FAST_P) | |
469 | PROFILE_COUNT_SCACHE_MISS (current_cpu); | |
470 | ||
471 | /* begin extract-scache */ | |
472 | EOF | |
473 | ||
474 | ${SHELL} $infile extract-scache | |
475 | ||
476 | cat << EOF | |
477 | /* end extract-scache */ | |
478 | } | |
479 | else if (FAST_P) | |
480 | { | |
481 | PROFILE_COUNT_SCACHE_HIT (current_cpu); | |
482 | /* Make core access statistics come out right. | |
483 | The size is a guess, but it's currently not used either. */ | |
484 | PROFILE_COUNT_CORE (current_cpu, vpc, 2, exec_map); | |
485 | } | |
486 | ||
487 | return sc; | |
488 | } | |
489 | ||
490 | #define FAST_P 0 | |
491 | ||
492 | void | |
493 | @cpu@_engine_run_full (SIM_CPU *current_cpu) | |
494 | { | |
495 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
496 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
497 | unsigned int hash_mask = CPU_SCACHE_HASH_MASK (current_cpu); | |
498 | SEM_PC vpc; | |
499 | ||
500 | EOF | |
501 | ||
502 | if [ x$parallel != xno ] ; then | |
503 | cat << EOF | |
504 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
505 | PAREXEC *par_exec; | |
506 | ||
507 | EOF | |
508 | fi | |
509 | ||
510 | # Any initialization code before looping starts. | |
511 | # Note that this code may declare some locals. | |
512 | ${SHELL} $infile init | |
513 | ||
514 | if [ x$parallel != xno ] ; then | |
515 | cat << EOF | |
516 | ||
517 | #if defined (HAVE_PARALLEL_EXEC) && defined (__GNUC__) | |
518 | { | |
519 | if (! CPU_IDESC_READ_INIT_P (current_cpu)) | |
520 | { | |
521 | /* ??? Later maybe paste read.c in when building mainloop.c. */ | |
522 | #define DEFINE_LABELS | |
523 | #include "readx.c" | |
524 | CPU_IDESC_READ_INIT_P (current_cpu) = 1; | |
525 | } | |
526 | } | |
527 | #endif | |
528 | ||
529 | EOF | |
530 | fi | |
531 | ||
532 | cat << EOF | |
533 | ||
534 | vpc = GET_H_PC (); | |
535 | ||
536 | do | |
537 | { | |
538 | SCACHE *sc; | |
539 | ||
540 | sc = @cpu@_scache_lookup (current_cpu, vpc, scache, hash_mask, FAST_P); | |
541 | ||
542 | /* begin full-exec-scache */ | |
543 | EOF | |
544 | ||
545 | ${SHELL} $infile full-exec-scache | |
546 | ||
547 | cat << EOF | |
548 | /* end full-exec-scache */ | |
549 | ||
550 | SET_H_PC (vpc); | |
551 | ||
552 | ++ CPU_INSN_COUNT (current_cpu); | |
553 | } | |
554 | while (0 /*CPU_RUNNING_P (current_cpu)*/); | |
555 | } | |
556 | ||
557 | #undef FAST_P | |
558 | ||
559 | EOF | |
560 | ||
561 | #################################### | |
562 | ||
563 | # Scache engine: fast version. | |
564 | ||
565 | if [ x$fast = xyes ] ; then | |
566 | ||
567 | cat << EOF | |
568 | ||
569 | #define FAST_P 1 | |
570 | ||
571 | void | |
572 | @cpu@_engine_run_fast (SIM_CPU *current_cpu) | |
573 | { | |
574 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
575 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
576 | unsigned int hash_mask = CPU_SCACHE_HASH_MASK (current_cpu); | |
577 | SEM_PC vpc; | |
578 | ||
579 | EOF | |
580 | ||
581 | if [ x$parallel != xno ] ; then | |
582 | cat << EOF | |
583 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
584 | PAREXEC *par_exec; | |
585 | ||
586 | EOF | |
587 | fi | |
588 | ||
589 | # Any initialization code before looping starts. | |
590 | # Note that this code may declare some locals. | |
591 | ${SHELL} $infile init | |
592 | ||
593 | if [ x$parallel != xno ] ; then | |
594 | cat << EOF | |
595 | ||
596 | #if defined (HAVE_PARALLEL_EXEC) && defined (__GNUC__) | |
597 | { | |
598 | if (! CPU_IDESC_READ_INIT_P (current_cpu)) | |
599 | { | |
600 | /* ??? Later maybe paste read.c in when building mainloop.c. */ | |
601 | #define DEFINE_LABELS | |
602 | #include "readx.c" | |
603 | CPU_IDESC_READ_INIT_P (current_cpu) = 1; | |
604 | } | |
605 | } | |
606 | #endif | |
607 | ||
608 | EOF | |
609 | fi # parallel != no | |
610 | ||
611 | cat << EOF | |
612 | ||
613 | #if WITH_SEM_SWITCH_FAST && defined (__GNUC__) | |
614 | { | |
615 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
616 | { | |
617 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
618 | #define DEFINE_LABELS | |
619 | #include "$switch" | |
620 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
621 | } | |
622 | } | |
623 | #endif | |
624 | ||
625 | vpc = GET_H_PC (); | |
626 | ||
627 | do | |
628 | { | |
629 | SCACHE *sc; | |
630 | ||
631 | sc = @cpu@_scache_lookup (current_cpu, vpc, scache, hash_mask, FAST_P); | |
632 | ||
633 | /* begin fast-exec-scache */ | |
634 | EOF | |
635 | ||
636 | ${SHELL} $infile fast-exec-scache | |
637 | ||
638 | cat << EOF | |
639 | /* end fast-exec-scache */ | |
640 | ||
641 | SET_H_PC (vpc); | |
642 | ||
643 | ++ CPU_INSN_COUNT (current_cpu); | |
644 | } | |
645 | while (0 /*CPU_RUNNING_P (current_cpu)*/); | |
646 | } | |
647 | ||
648 | #undef FAST_P | |
649 | ||
650 | EOF | |
651 | ||
652 | fi # -fast | |
653 | ||
654 | fi # -scache | |
655 | ||
656 | ########################################################################## | |
657 | ||
658 | # Compilation engine: lookup insn in scache, extract a pbb | |
659 | # (pseudo-basic-block) if missing, then execute the pbb. | |
660 | # A "pbb" is a sequence of insns up to the next cti insn or until | |
661 | # some prespecified maximum. | |
662 | # CTI: control transfer instruction. | |
663 | ||
664 | if [ x$pbb = xyes ] ; then | |
665 | ||
666 | cat << EOF | |
667 | ||
668 | /* Record address of cti terminating a pbb. */ | |
669 | #define SET_CTI_VPC(sc) do { _cti_sc = (sc); } while (0) | |
670 | /* Record number of [real] insns in pbb. */ | |
671 | #define SET_INSN_COUNT(n) do { _insn_count = (n); } while (0) | |
672 | ||
673 | /* Fetch and extract a pseudo-basic-block. | |
674 | FAST_P is non-zero if no tracing/profiling/etc. is wanted. */ | |
675 | ||
676 | INLINE SEM_PC | |
677 | @cpu@_pbb_begin (SIM_CPU *current_cpu, int FAST_P) | |
678 | { | |
679 | SEM_PC new_vpc; | |
680 | PCADDR pc; | |
681 | SCACHE *sc; | |
682 | int max_insns = CPU_SCACHE_MAX_CHAIN_LENGTH (current_cpu); | |
683 | ||
684 | pc = GET_H_PC (); | |
685 | ||
686 | new_vpc = scache_lookup_or_alloc (current_cpu, pc, max_insns, &sc); | |
687 | if (! new_vpc) | |
688 | { | |
689 | /* Leading '_' to avoid collision with mainloop.in. */ | |
690 | int _insn_count = 0; | |
691 | SCACHE *orig_sc = sc; | |
692 | SCACHE *_cti_sc = NULL; | |
693 | int slice_insns = CPU_MAX_SLICE_INSNS (current_cpu); | |
694 | ||
695 | /* First figure out how many instructions to compile. | |
696 | MAX_INSNS is the size of the allocated buffer, which includes space | |
697 | for before/after handlers if they're being used. | |
698 | SLICE_INSNS is the maxinum number of real insns that can be | |
699 | executed. Zero means "as many as we want". */ | |
700 | /* ??? max_insns is serving two incompatible roles. | |
701 | 1) Number of slots available in scache buffer. | |
702 | 2) Number of real insns to execute. | |
703 | They're incompatible because there are virtual insns emitted too | |
704 | (chain,cti-chain,before,after handlers). */ | |
705 | ||
706 | if (slice_insns == 1) | |
707 | { | |
708 | /* No need to worry about extra slots required for virtual insns | |
709 | and parallel exec support because MAX_CHAIN_LENGTH is | |
710 | guaranteed to be big enough to execute at least 1 insn! */ | |
711 | max_insns = 1; | |
712 | } | |
713 | else | |
714 | { | |
715 | /* Allow enough slop so that while compiling insns, if max_insns > 0 | |
716 | then there's guaranteed to be enough space to emit one real insn. | |
717 | MAX_CHAIN_LENGTH is typically much longer than | |
718 | the normal number of insns between cti's anyway. */ | |
719 | max_insns -= (1 /* one for the trailing chain insn */ | |
720 | + (FAST_P | |
721 | ? 0 | |
722 | : (1 + MAX_PARALLEL_INSNS) /* before+after */) | |
723 | + (MAX_PARALLEL_INSNS > 1 | |
724 | ? (MAX_PARALLEL_INSNS * 2) | |
725 | : 0)); | |
726 | ||
727 | /* Account for before/after handlers. */ | |
728 | if (! FAST_P) | |
729 | slice_insns *= 3; | |
730 | ||
731 | if (slice_insns > 0 | |
732 | && slice_insns < max_insns) | |
733 | max_insns = slice_insns; | |
734 | } | |
735 | ||
736 | new_vpc = sc; | |
737 | ||
738 | /* SC,PC must be updated to point passed the last entry used. | |
739 | SET_CTI_VPC must be called if pbb is terminated by a cti. | |
740 | SET_INSN_COUNT must be called to record number of real insns in | |
741 | pbb [could be computed by us of course, extra cpu but perhaps | |
742 | negligible enough]. */ | |
743 | ||
744 | /* begin extract-pbb */ | |
745 | EOF | |
746 | ||
747 | ${SHELL} $infile extract-pbb | |
748 | ||
749 | cat << EOF | |
750 | /* end extract-pbb */ | |
751 | ||
752 | /* The last one is a pseudo-insn to link to the next chain. | |
753 | It is also used to record the insn count for this chain. */ | |
754 | { | |
755 | const IDESC *id; | |
756 | ||
757 | /* Was pbb terminated by a cti? */ | |
758 | if (_cti_sc) | |
759 | { | |
760 | id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_CTI_CHAIN]; | |
761 | } | |
762 | else | |
763 | { | |
764 | id = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_CHAIN]; | |
765 | } | |
766 | SEM_SET_CODE (&sc->argbuf, id, FAST_P); | |
767 | sc->argbuf.idesc = id; | |
768 | sc->argbuf.addr = pc; | |
769 | sc->argbuf.fields.chain.insn_count = _insn_count; | |
770 | sc->argbuf.fields.chain.next = 0; | |
771 | ++sc; | |
772 | } | |
773 | ||
774 | /* Update the pointer to the next free entry. */ | |
775 | CPU_SCACHE_NEXT_FREE (current_cpu) = sc; | |
776 | /* Record length of chain if profiling. | |
777 | This includes virtual insns since they count against | |
778 | max_insns too. */ | |
779 | if (! FAST_P) | |
780 | PROFILE_COUNT_SCACHE_CHAIN_LENGTH (current_cpu, sc - orig_sc); | |
781 | } | |
782 | ||
783 | return new_vpc; | |
784 | } | |
785 | ||
786 | /* Chain to the next block from a non-cti terminated previous block. */ | |
787 | ||
788 | INLINE SEM_PC | |
789 | @cpu@_pbb_chain (SIM_CPU *current_cpu, SEM_ARG sem_arg) | |
790 | { | |
791 | ARGBUF *abuf = SEM_ARGBUF (sem_arg); | |
792 | ||
793 | PBB_UPDATE_INSN_COUNT (current_cpu, sem_arg); | |
794 | ||
795 | SET_H_PC (abuf->addr); | |
796 | ||
797 | /* If not running forever, exit back to main loop. */ | |
798 | if (CPU_MAX_SLICE_INSNS (current_cpu) != 0 | |
799 | /* Also exit back to main loop if there's an event. | |
800 | Note that if CPU_MAX_SLICE_INSNS != 1, events won't get processed | |
801 | at the "right" time, but then that was what was asked for. | |
802 | There is no silver bullet for simulator engines. | |
803 | ??? Clearly this needs a cleaner interface. | |
804 | At present it's just so Ctrl-C works. */ | |
805 | || STATE_EVENTS (CPU_STATE (current_cpu))->work_pending) | |
806 | CPU_RUNNING_P (current_cpu) = 0; | |
807 | ||
808 | /* If chained to next block, go straight to it. */ | |
809 | if (abuf->fields.chain.next) | |
810 | return abuf->fields.chain.next; | |
811 | /* See if next block has already been compiled. */ | |
812 | abuf->fields.chain.next = scache_lookup (current_cpu, abuf->addr); | |
813 | if (abuf->fields.chain.next) | |
814 | return abuf->fields.chain.next; | |
815 | /* Nope, so next insn is a virtual insn to invoke the compiler | |
816 | (begin a pbb). */ | |
817 | return CPU_SCACHE_PBB_BEGIN (current_cpu); | |
818 | } | |
819 | ||
820 | /* Chain to the next block from a cti terminated previous block. | |
821 | NEW_VPC_PTR is one of SEM_BRANCH_UNTAKEN, SEM_BRANCH_UNCACHEABLE, or | |
822 | a pointer to a location containing the SEM_PC of the branch's address. | |
823 | NEW_PC is the target's branch address, and is only valid if | |
824 | NEW_VPC_PTR != SEM_BRANCH_UNTAKEN. */ | |
825 | ||
826 | INLINE SEM_PC | |
827 | @cpu@_pbb_cti_chain (SIM_CPU *current_cpu, SEM_ARG sem_arg, | |
828 | SEM_PC *new_vpc_ptr, PCADDR new_pc) | |
829 | { | |
830 | ARGBUF *abuf; | |
831 | ||
832 | PBB_UPDATE_INSN_COUNT (current_cpu, sem_arg); | |
833 | ||
834 | /* If not running forever, exit back to main loop. */ | |
835 | if (CPU_MAX_SLICE_INSNS (current_cpu) != 0 | |
836 | /* Also exit back to main loop if there's an event. | |
837 | Note that if CPU_MAX_SLICE_INSNS != 1, events won't get processed | |
838 | at the "right" time, but then that was what was asked for. | |
839 | There is no silver bullet for simulator engines. | |
840 | ??? Clearly this needs a cleaner interface. | |
841 | At present it's just so Ctrl-C works. */ | |
842 | || STATE_EVENTS (CPU_STATE (current_cpu))->work_pending) | |
843 | CPU_RUNNING_P (current_cpu) = 0; | |
844 | ||
845 | /* Restart compiler if we branched to an uncacheable address | |
846 | (e.g. "j reg"). */ | |
847 | if (new_vpc_ptr == SEM_BRANCH_UNCACHEABLE) | |
848 | { | |
849 | SET_H_PC (new_pc); | |
850 | return CPU_SCACHE_PBB_BEGIN (current_cpu); | |
851 | } | |
852 | ||
853 | /* If branch wasn't taken, update the pc and set BR_ADDR_PTR to our | |
854 | next chain ptr. */ | |
855 | if (new_vpc_ptr == SEM_BRANCH_UNTAKEN) | |
856 | { | |
857 | abuf = SEM_ARGBUF (sem_arg); | |
858 | SET_H_PC (abuf->addr); | |
859 | new_vpc_ptr = &abuf->fields.chain.next; | |
860 | } | |
861 | else | |
862 | { | |
863 | SET_H_PC (new_pc); | |
864 | } | |
865 | ||
866 | /* If chained to next block, go straight to it. */ | |
867 | if (*new_vpc_ptr) | |
868 | return *new_vpc_ptr; | |
869 | /* See if next block has already been compiled. */ | |
870 | *new_vpc_ptr = scache_lookup (current_cpu, GET_H_PC ()); | |
871 | if (*new_vpc_ptr) | |
872 | return *new_vpc_ptr; | |
873 | /* Nope, so next insn is a virtual insn to invoke the compiler | |
874 | (begin a pbb). */ | |
875 | return CPU_SCACHE_PBB_BEGIN (current_cpu); | |
876 | } | |
877 | ||
878 | /* x-before handler. | |
879 | This is called before each insn. */ | |
880 | ||
881 | void | |
882 | @cpu@_pbb_before (SIM_CPU *current_cpu, SCACHE *sc) | |
883 | { | |
884 | SEM_ARG sem_arg = sc; | |
885 | const ARGBUF *abuf = SEM_ARGBUF (sem_arg); | |
886 | int first_p = abuf->fields.before.first_p; | |
887 | const ARGBUF *cur_abuf = SEM_ARGBUF (sc + 1); | |
888 | const IDESC *cur_idesc = cur_abuf->idesc; | |
889 | PCADDR pc = cur_abuf->addr; | |
890 | ||
891 | if (ARGBUF_PROFILE_P (cur_abuf)) | |
892 | PROFILE_COUNT_INSN (current_cpu, pc, cur_idesc->num); | |
893 | ||
894 | /* If this isn't the first insn, finish up the previous one. */ | |
895 | ||
896 | if (! first_p) | |
897 | { | |
898 | if (PROFILE_MODEL_P (current_cpu)) | |
899 | { | |
900 | const SEM_ARG prev_sem_arg = sc - 1; | |
901 | const ARGBUF *prev_abuf = SEM_ARGBUF (prev_sem_arg); | |
902 | const IDESC *prev_idesc = prev_abuf->idesc; | |
903 | int cycles; | |
904 | ||
905 | /* ??? May want to measure all insns if doing insn tracing. */ | |
906 | if (ARGBUF_PROFILE_P (prev_abuf)) | |
907 | { | |
908 | cycles = (*prev_idesc->timing->model_fn) (current_cpu, prev_sem_arg); | |
909 | @cpu@_model_insn_after (current_cpu, 0 /*last_p*/, cycles); | |
910 | } | |
911 | } | |
912 | ||
913 | TRACE_INSN_FINI (current_cpu, cur_abuf, 0 /*last_p*/); | |
914 | } | |
915 | ||
916 | /* FIXME: Later make cover macros: PROFILE_INSN_{INIT,FINI}. */ | |
917 | if (PROFILE_MODEL_P (current_cpu) | |
918 | && ARGBUF_PROFILE_P (cur_abuf)) | |
919 | @cpu@_model_insn_before (current_cpu, first_p); | |
920 | ||
921 | TRACE_INSN_INIT (current_cpu, cur_abuf, first_p); | |
922 | TRACE_INSN (current_cpu, cur_idesc->idata, cur_abuf, pc); | |
923 | } | |
924 | ||
925 | /* x-after handler. | |
926 | This is called after a serial insn or at the end of a group of parallel | |
927 | insns. */ | |
928 | ||
929 | void | |
930 | @cpu@_pbb_after (SIM_CPU *current_cpu, SCACHE *sc) | |
931 | { | |
932 | SEM_ARG sem_arg = sc; | |
933 | const ARGBUF *abuf = SEM_ARGBUF (sem_arg); | |
934 | const SEM_ARG prev_sem_arg = sc - 1; | |
935 | const ARGBUF *prev_abuf = SEM_ARGBUF (prev_sem_arg); | |
936 | ||
937 | /* ??? May want to measure all insns if doing insn tracing. */ | |
938 | if (PROFILE_MODEL_P (current_cpu) | |
939 | && ARGBUF_PROFILE_P (prev_abuf)) | |
940 | { | |
941 | const IDESC *prev_idesc = prev_abuf->idesc; | |
942 | int cycles; | |
943 | ||
944 | cycles = (*prev_idesc->timing->model_fn) (current_cpu, prev_sem_arg); | |
945 | @cpu@_model_insn_after (current_cpu, 1 /*last_p*/, cycles); | |
946 | } | |
947 | TRACE_INSN_FINI (current_cpu, prev_abuf, 1 /*last_p*/); | |
948 | } | |
949 | ||
950 | #define FAST_P 0 | |
951 | ||
952 | void | |
953 | @cpu@_engine_run_full (SIM_CPU *current_cpu) | |
954 | { | |
955 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
956 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
957 | /* virtual program counter */ | |
958 | SEM_PC vpc; | |
959 | #if WITH_SEM_SWITCH_FULL | |
960 | /* For communication between cti's and cti-chain. */ | |
961 | PCADDR pbb_br_npc; | |
962 | SEM_PC *pbb_br_npc_ptr; | |
963 | #endif | |
964 | ||
965 | EOF | |
966 | ||
967 | if [ x$parallel != xno ] ; then | |
968 | cat << EOF | |
969 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
970 | PAREXEC *par_exec = &pbufs[0]; | |
971 | ||
972 | EOF | |
973 | fi | |
974 | ||
975 | # Any initialization code before looping starts. | |
976 | # Note that this code may declare some locals. | |
977 | ${SHELL} $infile init | |
978 | ||
979 | cat << EOF | |
980 | ||
981 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
982 | { | |
983 | /* ??? 'twould be nice to move this up a level and only call it once. | |
984 | On the other hand, in the "let's go fast" case the test is only done | |
985 | once per pbb (since we only return to the main loop at the end of | |
986 | a pbb). And in the "let's run until we're done" case we don't return | |
987 | until the program exits. */ | |
988 | ||
989 | #if WITH_SEM_SWITCH_FULL && defined (__GNUC__) | |
990 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
991 | #define DEFINE_LABELS | |
992 | #include "$switch" | |
993 | #endif | |
994 | ||
995 | /* Initialize the "begin (compile) a pbb" virtual insn. */ | |
996 | vpc = CPU_SCACHE_PBB_BEGIN (current_cpu); | |
997 | SEM_SET_FULL_CODE (SEM_ARGBUF (vpc), | |
998 | & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]); | |
999 | vpc->argbuf.idesc = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]; | |
1000 | ||
1001 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
1002 | } | |
1003 | ||
1004 | CPU_RUNNING_P (current_cpu) = 1; | |
1005 | /* ??? In the case where we're returning to the main loop after every | |
1006 | pbb we don't want to call pbb_begin each time (which hashes on the pc | |
1007 | and does a table lookup). A way to speed this up is to save vpc | |
1008 | between calls. */ | |
1009 | vpc = @cpu@_pbb_begin (current_cpu, FAST_P); | |
1010 | ||
1011 | do | |
1012 | { | |
1013 | /* begin full-exec-pbb */ | |
1014 | EOF | |
1015 | ||
1016 | ${SHELL} $infile full-exec-pbb | |
1017 | ||
1018 | cat << EOF | |
1019 | /* end full-exec-pbb */ | |
1020 | } | |
1021 | while (CPU_RUNNING_P (current_cpu)); | |
1022 | } | |
1023 | ||
1024 | #undef FAST_P | |
1025 | ||
1026 | EOF | |
1027 | ||
1028 | #################################### | |
1029 | ||
1030 | # Compile engine: fast version. | |
1031 | ||
1032 | if [ x$fast = xyes ] ; then | |
1033 | ||
1034 | cat << EOF | |
1035 | ||
1036 | #define FAST_P 1 | |
1037 | ||
1038 | void | |
1039 | @cpu@_engine_run_fast (SIM_CPU *current_cpu) | |
1040 | { | |
1041 | SIM_DESC current_state = CPU_STATE (current_cpu); | |
1042 | SCACHE *scache = CPU_SCACHE_CACHE (current_cpu); | |
1043 | /* virtual program counter */ | |
1044 | SEM_PC vpc; | |
1045 | #if WITH_SEM_SWITCH_FAST | |
1046 | /* For communication between cti's and cti-chain. */ | |
1047 | PCADDR pbb_br_npc; | |
1048 | SEM_PC *pbb_br_npc_ptr; | |
1049 | #endif | |
1050 | ||
1051 | EOF | |
1052 | ||
1053 | if [ x$parallel != xno ] ; then | |
1054 | cat << EOF | |
1055 | PAREXEC pbufs[MAX_PARALLEL_INSNS]; | |
1056 | PAREXEC *par_exec = &pbufs[0]; | |
1057 | ||
1058 | EOF | |
1059 | fi | |
1060 | ||
1061 | # Any initialization code before looping starts. | |
1062 | # Note that this code may declare some locals. | |
1063 | ${SHELL} $infile init | |
1064 | ||
1065 | cat << EOF | |
1066 | ||
1067 | if (! CPU_IDESC_SEM_INIT_P (current_cpu)) | |
1068 | { | |
1069 | /* ??? 'twould be nice to move this up a level and only call it once. | |
1070 | On the other hand, in the "let's go fast" case the test is only done | |
1071 | once per pbb (since we only return to the main loop at the end of | |
1072 | a pbb). And in the "let's run until we're done" case we don't return | |
1073 | until the program exits. */ | |
1074 | ||
1075 | #if WITH_SEM_SWITCH_FAST && defined (__GNUC__) | |
1076 | /* ??? Later maybe paste sem-switch.c in when building mainloop.c. */ | |
1077 | #define DEFINE_LABELS | |
1078 | #include "$switch" | |
1079 | #endif | |
1080 | ||
1081 | /* Initialize the "begin (compile) a pbb" virtual insn. */ | |
1082 | vpc = CPU_SCACHE_PBB_BEGIN (current_cpu); | |
1083 | SEM_SET_FAST_CODE (SEM_ARGBUF (vpc), | |
1084 | & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]); | |
1085 | vpc->argbuf.idesc = & CPU_IDESC (current_cpu) [@CPU@_INSN_X_BEGIN]; | |
1086 | ||
1087 | CPU_IDESC_SEM_INIT_P (current_cpu) = 1; | |
1088 | } | |
1089 | ||
1090 | CPU_RUNNING_P (current_cpu) = 1; | |
1091 | /* ??? In the case where we're returning to the main loop after every | |
1092 | pbb we don't want to call pbb_begin each time (which hashes on the pc | |
1093 | and does a table lookup). A way to speed this up is to save vpc | |
1094 | between calls. */ | |
1095 | vpc = @cpu@_pbb_begin (current_cpu, FAST_P); | |
1096 | ||
1097 | do | |
1098 | { | |
1099 | /* begin fast-exec-pbb */ | |
1100 | EOF | |
1101 | ||
1102 | ${SHELL} $infile fast-exec-pbb | |
1103 | ||
1104 | cat << EOF | |
1105 | /* end fast-exec-pbb */ | |
1106 | } | |
1107 | while (CPU_RUNNING_P (current_cpu)); | |
1108 | } | |
1109 | ||
1110 | #undef FAST_P | |
1111 | ||
1112 | EOF | |
1113 | fi # -fast | |
1114 | ||
1115 | fi # -pbb | |
1116 | ||
1117 | # Process @cpu@,@CPU@ appearing in mainloop.in. | |
1118 | sed -e "s/@cpu@/$cpu/g" -e "s/@CPU@/$CPU/g" < tmp-mloop.cin > mloop.cin | |
1119 | rc=$? | |
1120 | rm -f tmp-mloop.cin | |
1121 | ||
1122 | exit $rc |